prediction-pillar

online state construction

This repository contains implementation for:

Vanilla RNN/ GRU learned semi-online using Truncated Backprop

Trace RNN implementation: Some hidden units of the RNN are trace features

Recurrent Geoff Task

Online Mackeyglass and MSO time series

There are also these experiments in the experiment directory:

learning state in cycle world with trace features

delayed mackeyglass prediction

episodic, continuing recurrent geoff task

BPTT performance on non-stationary geoff task experiment